Release decision
The team chose weekly releases after the launch review.
- Tags
- release-plan
- Origin
- Launch review
Original sourceLaunch review
Attached to this recordMemory Observatory
Observatory
Open the record behind the answer.
Keep the source within reach.
Keep the original source attached so your team can check the information behind a decision.
Illustrative record inspector. These are sample memories, not live operational data.
How we evaluate memory
A useful answer starts with the right context.
Memory evaluation asks two different questions: did retrieval find the relevant information, and did the agent use it to answer correctly? We examine both when developing Akasha, the engine behind Minds.
Our internal LongMemEval work includes checks for reference answers within retrieved context. Those checks measure evidence coverage. An answer-quality comparison also needs the dataset version, retrieval limits, model, judge, and scoring method to be clear.
Discuss evaluation for your workloadThe platform at a glance
Built around the way your agents work.
0
kinds of memory
Experiences, workflows, resources, secrets, working context, and patterns.
0
ways to work with data
Records, vector search, a knowledge graph, and SQL analytics in one engine.
0
dedicated Mind
Your own instance, with the same Mind when you upgrade from Free to Pro.
0
ways to get started
Talk to Helm, connect an SDK, or work from the command line.